Problem with high pitch noises

Information on strange high pitch, low volume noises emitted by ThinkPads.

Problem description

Even though ThinkPads are known as very silent notebooks, they tend to emit different, mostly high pitch noises in certain circumstances. The noises are of low volume and hence not realized by everyone or at least tolerated by most people. However, there are those with "bat like ears" that hear them and might be annoyed by that.

Affected Models

Noises have been experienced in the following situations:

situation	noise description	affected models
Plugged into AC / running at high CPU frequency	soft crackling, buzzing noise	T41 2379-DJU
ThinkPad suspended to RAM	constant high pitch noise	T42p
ThinkPad connected to power and switched off, with battery fully charged	constant high pitch noise	T41p 2373-GHG
moving windows or just the mouse in xorg	strange noise like a rapid series of very short high pitch noises adding to a constant kind of whistling (only as long as the movement goes)	A21m 2628 T41p 2373-GHG
Thinkpad connected to power or working on battery, also when suspended to RAM	High pitch noise also when HD is powered down.	T23 2647-DG4 X24 2662-MWG
Thinkpad connected to power battery charged less than 60%	High pitch noise till battery is charged more than 60%.	T41p 2373-GEG
Constantly, if AC connected	High pitched, low volume constant noise.	T40 2373-88U
When the CPU freq jumps up to 1Ghz or above, or when the laptop is in suspend mode	Constant high pitched	R50 1829-6DM
(May come from harddisk.)	continuous, intermittent, low volume, high pitched	600X
ACPI puts the processor into the C3 or C4 power saving states (i.e., the system is idle).	High-pitched crackling noise.	X41 T43 2686-DGU T43p T60 2007-72U T21 T20 2648-46U
Cpufreqd (or powernowd etc.) slows down the processor, for exemple at the end of an heavy task.	High-pitched crackling noise.	T42
When on battery	Noise level varies from inaudible to clearly audible, screeching sound, or sometimes beeping (when wifi is on).	A21m 2628 R51 EHG-1829 R52 1858-A11 1846-B5G R60 9461-DXG 9462-77G 9456-HTG 9461-DXG R60e 0657-A9G 0657-3LG T43 2686-DGU 1871-4AG T60 2007-72U 1951-24G 1951-24G 2007-49G 2007-FUG T60p 2007-FBG X60 X31 2673-CBU Z61m 9450-3HG 9452-128 9452-W5Q Z61t 9440-2QU
CPU is used much AND laptop is on AC power AND the TFT panel is enabled.	Screetching high pitch noise, like a million crickets	X41
Network is pluged in and networkload is ~100%.	low volume highfreq. pitch noise from left speaker or cpu.	Z60m, Z60t T40 T43 2687-DDU

Affected Operating Systems

All, though Linux appears especially susceptible.

Possible sources

• CPU activity: On some models the noise is triggered by certain CPU power states or activity patterns (as proven here).
• Graphics processor: In some models from the T2x era, e.g. the T23, the problem was related to the graphics circuitry and occured especially or only while making use of DirectDraw functions. IBM was able to fix it through a BIOS upgrade.
• Hard disk: On some ThinkPads the processor and hard disk are adjacent and produce similar noise. For example, in the X41 the sound generated by the hard disk is likely to be loudest at the vent.
• Gigabit Ethernet processor: In some models (e.g., T43) high ethernet activity causes a high pitch noise.

• Screen brightness: on an X31, a hissing sound is started whenever screen brightness is not full.

Status

The problem is highly specific to operating system, model and even individual machines. There is no universal solution, but on most machines one of the following will reduce or eliminate the noise (possibly at some cost in power consumption).

Solutions for CPU-triggered noise

Limit ACPI CPU power states

There are eight ACPI CPU power states, called C1 through C8 (but the ThinkPad BIOS often remaps them and uses only C1 to C4). Often only the extreme power saving modes C4 or C3 produce the noise, so the noise can be stopped by instructing the Linux ACPI code to use only lower modes.

To forbid the ACPI driver from using C4 (this fixed the problem with on some ThinkPad T43, T43p and T41):

• If the ACPI processor component is compiled as built-in (CONFIG_ACPI_PROCESSOR=y):
  • Pass the max_cstate=3 kernel argument.
• If the ACPI processor component is loaded as a module (CONFIG_ACPI_PROCESSOR=m and processor shows in the output of # lsmod), do either of:
  • Pass the processor.max_cstate=3 kernel argument (this does not work in Ubuntu 5.10 with default kernel).
  • Add options processor max_cstate=3 to /etc/modprobe.conf (or /etc/modprobe.conf.local, or /etc/modprobe.d/..., depending on your system) (this does not work in Ubuntu 5.10 with default kernel).
  • # echo 3 > /sys/module/processor/parameters/max_cstate (this can be changed in runtime for experimentation). (If may need to be set again upon resume from suspend, e.g., in the wakeup script.)
  • (On Ubuntu 5.10, the default kernel uses processor as a module. Unfortunately, the script loading it, /etc/init.d/acpid, ignores the options processor max_cstate=3 setting in /etc/modprobe.d/<my file>. As a solution for this specific problem, add the line echo 2 > /sys/module/processor/parameters/max_cstate directly to /etc/init.d/acpid, at the end of the function load_modules(), immediately after the line echo "$PRINTK" > /proc/sys/kernel/printk.)
  • On Gentoo: Configure your Thinkpad as described in the Power Management Guide. Then as root create the files /etc/init.d/limit-sleep-states and /etc/conf.d/limit-sleep-states as below. After that just issue the command # rc-update add limit-sleep-states default to limit the sleep states only when running on AC power, or # rc-update add limit-sleep-states default battery to always limit the sleep states.

cat <<EOF > /etc/init.d/limit-sleep-states
#!/sbin/runscript

depend() {
	need acpid
}

start() {
	ebegin "Limiting CPU sleep state to C${LIMIT_CSTATE}"
	echo $LIMIT_CSTATE > /sys/module/processor/parameters/max_cstate
	eend $?
}

stop() {
       ebegin "Removing CPU sleep state limit"
       echo $REMOVE_CSTATE > /sys/module/processor/parameters/max_cstate
       eend $?
}
EOF

cat <<EOF > /etc/conf.d/limit-sleep-states
# limit CPU sleep state to the following value (adjust accordingly)
LIMIT_CSTATE=3
# some value which is higher than all available sleep states
REMOVE_CSTATE=8
EOF

To also forbid the C3 state, replace "3" with "2" above (this fixed the problem on some ThinkPad X40, X41, X60, T60, Z61t and R52, as well as on T20 where C4 is not supported at all).

Note that these options affect power consumption when the CPU is idle. For example, here is the measured power consumption on a ThinkPad T43:

• processor.max_cstate=4: 15160mW (default, noisy)
• processor.max_cstate=3: 15770mW (660mW higher, silent)
• processor.max_cstate=2: 16100mW (2940mW higher, silent)

One can control how often Linux tries to enter the C3 state by using the processor.bm_history=<bitmask> parameter on kernels with a scheduler frequency below 800Hz (and if you have noise problems, you really should not be running the kernel at 1000Hz...). Setting processor.bm_history=0xFFFFFFFF will cause C3 to be entered less often. This will waste more power as the CPU won't do C3 or C4 as often, but at least it doesn't forbid C3 and C4 permanently, unlike max_cstate=2.

See the discussion page for further information and success reports.

• Jakob Schou Pedersen: Editing the file /etc/init.d/acpid as described above (the last solution) worked on my T43 :-)

Turn off CPU power saving in the BIOS

Go into the BIOS and turn off the power saving processor feature that puts it into idle mode. (This worked on a ThinkPad T43, T21, X60s and Z61m). However, this also affects power consumption when the CPU is idle, it's similar to disabling the C4/C3 ACPI CPU power state. For example, on a Z61m 9450-3HG, a full charged battery with power management enabled in the BIOS provides the notebook with power for about 3.5h, if disabled, the battery-lifetime is only about 2h.

Disable ACPI CPU power states

Completely disable CPU ACPI power states. Discussion:

• From Martin Steigerwald: I made the observation that I get at least less high pitch noises on my T23 when I do not use the two ACPI modules "processor" and "thermal" (depends on the first one). I have no clue, why. Anyone with similar experiences?

Omar Yasin: I've got a R52 and when I load the same ACPIO modules the high pitch noises are not as loud but I can still hear them.

• Kaspar Schleiser: On my T23 the noise is less loud when setting max_cstate to 2, but to completely silence it, I have to set it to "1". Is that the same as removing the "processor"-module? "cat /proc/acpi/battery/BAT0/state" does not show increased power drain.
• Niko Ehrenfeuchter: I'm experiencing the same here on my X24. Removing the "processor" module also stops the pitch noise, which does ONLY occur when setting the CPU to maximum speed (using cpufreq). On low speed it's completely silent, even having loaded the processor module.

• Rolf Adelsberger: I can confirm this: the high pitch noise is only remarkable (at least with my ears ;-) ) if the processor speed is set to maximum frequency.

• Stefan Baums: My X41 produced a high-pitched crackle from the processor vent on the left. Changing HZ did nothing, and the 'processor' module could not easily be removed from the system (Ubuntu 5.04). What solved the problem for me was adding idle=halt to the boot command line. Unfortunately, this solution only lasts until the first hibernation or suspend - when the computer (X41) resumes, the high-pitched crackle is back.

• The idle=halt solution combined with setting #define HZ 100 in the kernel fixes the problem on a T43.

• jhatch: idle=halt plus #define HZ 100 also worked on my T43. It still reverts back to noisy after a suspend/resume though. This needs to be fixed...

Change the timer interrupt frequency

Change the "HZ" kernel constants to alter the frequency of timer interrupts. Discussion:

NOTE!

The timer interrupt frequency (HZ) in current Linux kernels is directly tied to the kernel task scheduler. Lower frequencies provide larger time-slices and thus also higher latencies (which may kill latency-sensitive applications like audio processing). 100Hz ended up as the recommended "server" setting (because it increases disk/CPU throughput in a latency-insensitive environment). Higher frequencies are better for latency-sensitive applications, and improve desktop responsivity at the cost of less processor throughput. 1000Hz ended up as the recommended "desktop" setting.

• Andreas Karnahl: i've read in several forums it has something to do with the "idle"-state (or "C3") of the processor. There is a frequency called "timer interrupt" (or so mething like that). Since kernel 2.6x it is set to 1000 Hz by default (compared to 100 Hz in Kernel 2.4x). The exact reason i don't know, but it is safe to change this frequency to 100 Hz in kernel 2.6x (by the way, windows up to XP uses 100 Hz by default).
  Just do the following:
  

In [path to kernel-sources]/include/asm-i386/param.h find the line

#define HZ 1000

and change the value of HZ to 100:

#define HZ 100

Then recompile the kernel.
After i changed it on my ThinkPad A30 (under SuSE 9.2 and 9.3) and recompiling the kernel the high pitch noise is gone away.

• Omar Yasin: Worked on my R52, thanks.

• Thinker: In modern kernels this constant is in the kernel configuration Timer Frequency → Processor type and features (CONFIG_HZ).

Prevent idling

Indirectly avoid power saving states by making sure the CPU is rarely idle:

• Paul RIVIER: Here is a really simple workaround. C3 / C4 states are mainly called when the cpu freq is higher than required, for example if your cpufreqd is lazy to slow down the frequency but quick to raise it. That is why I use powernowd with the builtin "passive" mode, which is lazy for raising frequency, but quick to go back to the lowest. Now I don't hear them as often as before, as I avoid C3/C4 states at high frequency.

• The problem also occurs on my X41 with 2.6.11. Setting up frequency scaling with the ondemand governor makes things a lot better, as the processor does not stay with the maximum frequency when in idle mode. It can be still heard sometimes, though.

• On a T43 the noise was gone after dropping cpufreqd and switching to the ondemand governor - maybe because of the high sampling rate? (used the default: 10ms)

Change the processor voltage

Reducing the processor voltage (when possible) may decrease or eliminate the noise. On one ThinkPad T43, undervolting the Pentium M processor eliminated the high-pitched noise. Compared to the other solutions this has the benefit of lower power consumption, both due to the undervolting itself and because there is no need to forbid high APCI CPU power saving modes.

Solutions for screen brightness related sounds

Disable BIOS brightness control

The sound starts when on batteries as the BIOS automatically reduces screen brightness. To disable this, simply switch Config>Display>Brightness from Normal to High. See also here.

Other solutions

Disable UltraBay

• Naheed Vora: My T41 (2373-268) started to give high pitch noise ocassionally, when I upgraded to 2.6.11 kernel. I tried to unload lot of modules but finally figured out that disabling bay stops the noise. If you have ibm-acpi, do (need a cleaner solution): $ echo eject >/proc/acpi/ibm/bay .

Disable IrDA

• Mike Perry: I was able to cure an intermittent high-pitched whine on both my X24 and X40 by disabling the Infrared port.

Disable the Linuxant Modem Driver

• Joern Heissler: I made another experience. I played around with linuxant conexant modem drivers. After loading them I got some noise on my T42p.

Media Player paused

• Eilif Muller: On my R52 the high-pitched noises go away if I load XMMS, play something then pause it.
• Jacob: On my T43 DGU it goes away if I open mplayerc.exe and press play then pause it. This is the high-pitched noise that only shows up when I'm on battery.

I'd guess that the above tip works as when xmms or similar is running, it is uncompressing compressed audio/video, which is a processor intensive action. Keeping this paused means that the app won't 'let go' of the processor, forcing it to stay up and running, which stops it entering the higher powersave modes.

nice yes

Run the command:

$ nice yes > /dev/null.

This is a good way to test whether the processor is causing the interference, since it forces the CPU to stay at full power. Of course, this will make your system get warm, and probably turn on the fan, as well as eating power.

Upgrade BIOS

On a ThinkPad X60s, upgrading the BIOS to version 1.06 eliminated the high pitch noise when running on battery.

Disable USB

On one T43 and one X60, (partially) disabling USB using # rmmod uhci_hcd significantly reduced the noise. Also, on a T60 this trick worked well, doing rmmod/modprobe twice wiped the noise permanently.

This is likely a CPU-induced noise case, as disabling USB UHCI will reduce a lot the amount of busmaster activity while the computer is idle, and thus has a direct effect on the need for ACPI C-state transitions from C3/C4 to C2 or lower states in a T43 or X60.